Abnormality identification method for high-voltage box of energy storage system

Abstract

The invention relates to the technical field of energy storage systems, in particular to an energy storage system high-voltage box abnormity identification method, which comprises the following steps of S1, acquiring multi-source data in each battery cluster in an energy storage system; and S2, according to a data preprocessing principle, for abnormal data exceeding a normal range in the multi-source data, when the method is used, the internal abnormity of the high-voltage box is adaptively and rapidly identified through an abnormity identification algorithm, meanwhile, early warning is performed on an abnormal change trend, data features are scored, and the abnormal data in the high-voltage box is acquired. The abnormal degree of the sample point in the energy storage system is quantized, and the working efficiency of the abnormal degree of the quantized sample is improved; through an anomaly recognition algorithm, the abnormal condition in the high-voltage box of the energy storage system battery cluster is adaptively positioned, and then abnormal outlier sample points are adaptively recognized without priori knowledge; and meanwhile, according to the change slope of the scoring result, the abnormal change trend of each battery cluster is quantified and early warning is performed in time, so that the safety and the use performance of the energy storage system are improved.

Description

Technical Field

[0001] This invention relates to the field of energy storage system technology, and more specifically to a method for identifying anomalies in the high-voltage box of an energy storage system. Background Technology

[0002] Within an energy storage system, the high-voltage box is a crucial device connecting the power circuit system (PCS) and the battery clusters. It contains key components of the main high-voltage circuit and is essential for the efficient and safe operation of the energy storage system. Abnormalities within the high-voltage box can cause battery cluster shutdowns, and in severe cases, thermal runaway. Therefore, alarm and early warning systems for abnormalities within the high-voltage box are extremely important.

[0003] Prior art 1: Patent publication number CN118112448A describes in its specification that "This disclosure provides a method, device, equipment, and storage medium for early warning of faults in vulnerable components of an energy storage system. The method includes: acquiring performance parameters of a battery pack; if the performance parameters of the battery pack meet preset fault conditions, determining the number of times the performance parameters of the battery pack meet the preset fault conditions; and obtaining early warning information for vulnerable components based on the number of faults and the preset early warning conditions, thereby providing early warning of faults in vulnerable components. This disclosure achieves a risk warning function for vulnerable components of the battery pack by analyzing and recording the amplitude and frequency of transiently acquired data, thus improving..." While the aforementioned technology aims to "improve fault early warning efficiency and achieve high fault location accuracy," it relies on prior knowledge, preset early warning thresholds, and preset fault counts for alarms. It cannot adaptively identify anomalies and requires using the cumulative number of faults over a period of time as a severity criterion for low, medium, and high-risk warnings, thus reducing the efficiency of quantifying sample anomalies. This results in the inefficiency of the technology, which aims to improve fault early warning efficiency and achieve high fault location accuracy.

[0004] Prior Art 2: Patent publication number CN118275872A describes in its specification that "This disclosure provides a method, apparatus, device, vehicle, and storage medium for detecting the state of a fuse. This disclosure relates to the field of vehicle technology. In some embodiments of this disclosure, under the target charging mode of the vehicle, the target temperature rise of the fuse within a set time period is monitored. Based on the target charging mode and the initial temperature, a mapping relationship table between the charging mode, the initial temperature range, and the temperature rise threshold is consulted to obtain the first temperature rise threshold corresponding to the vehicle. Based on the target temperature rise and the first temperature rise threshold, the state of the fuse is determined, and the mapping relationship table is matched one-to-one with the vehicle." In other words, each vehicle corresponds to a mapping table. Since the internal resistance of fuses differs between different vehicles, each vehicle compares its own data with the data in its own mapping table, rather than comparing the data with the data of other vehicles. This eliminates the difference in the internal resistance of fuses between different vehicles and results in high recognition accuracy. Although the above technology obtains the first temperature rise threshold corresponding to the vehicle by querying the mapping table between charging mode, initial temperature range and temperature rise threshold, and determines the status of the fuse, the above technology requires the construction of a mapping table for data under different charging modes of each vehicle to identify abnormal states, resulting in the need for a large amount of basic vehicle data.

[0005] In conclusion, developing a method for identifying anomalies in the high-voltage box of an energy storage system remains a critical issue that urgently needs to be addressed in the field of energy storage system technology. Summary of the Invention

[0006] The purpose of this invention is to solve the problems in existing technologies, which rely on prior knowledge, preset warning condition thresholds, and preset fault counts for alarms, and cannot adaptively identify anomalies; they require the cumulative number of faults over a period of time as the severity for low, medium, and high risk warnings, which reduces the efficiency of quantifying the degree of anomalies in samples; and they require the construction of a mapping table for data under different charging modes for each vehicle to identify abnormal states, which requires a large amount of basic vehicle data.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] This invention provides a method for identifying anomalies in high-voltage boxes of an energy storage system, comprising the following steps: S1, acquiring multi-source data from each battery cluster within the energy storage system;

[0009] S2. Based on the data preprocessing principles, the abnormal data exceeding the normal range in the multi-source data are cleaned and filled to obtain current information;

[0010] S3. Determine the operating status of the battery cluster based on the current information and output the determination result;

[0011] S4. Based on the judgment result, obtain the data segment within the charging and discharging time period, perform principal component analysis on the feature values ​​collected from the data segment, and extract data features;

[0012] S5. Input the data features into the anomaly identification algorithm to calculate the degree of anomaly of each high-voltage box in the battery cluster, identify high-voltage boxes with abnormal slopes, and issue an anomaly warning.

[0013] Furthermore, in step S1, the method for obtaining multi-source data within each battery cluster of the energy storage system is as follows:

[0014] The multi-source data includes timestamps within each battery cluster, total current, main positive and main negative terminals within the high-voltage box, pre-charge relay status, fuse temperature within the high-voltage box, connector insertion temperature, and measured voltage within the high-voltage box for each battery cluster. Set the collection period At every moment Collect the following multi-source data ,expression:

[0015] ,

[0016] In the formula, Indicates the first Individual battery clusters at a given time point The collected multi-source data set Indicates the first The timestamp of each sampling time point It is a timestamp indicating the number The specific time of each data collection Indicates the first Individual battery clusters in time Total current, Indicates the first Individual battery clusters in time The state of the main positive relay, Indicates the first Individual battery clusters in time The status of the main negative relay, Indicates the first Individual battery clusters in time The status of the pre-charged relay, Indicates the first Individual battery clusters in time The temperature of the fuse, Indicates the first Individual battery clusters in time The temperature of the connector insertion point Indicates the first Individual battery clusters in time The voltage measured inside the high-voltage box.

[0017] Furthermore, in step S2, the abnormal data exceeding the normal range in the multi-source data is cleaned and filled according to data preprocessing principles to obtain the current information. The method is as follows:

[0018] The cleaning and filling process involves removing duplicate data, using a set mapping function to uniquely process the timestamp set, filling missing values ​​with adjacent values ​​using a linear interpolation filling mechanism, and constructing a joint Gaussian model for abnormal current, temperature, and voltage data caused by any acquisition or communication issues. Mahalanobis distance is defined to detect anomalies, and Fourier transform is used to detect frequency drift for filtering. The expression is:

[0019] ,

[0020] In the formula, This represents the Fourier transform operator. Indicates the first Individual battery clusters in time The current signal, The integral symbol represents the integral from negative infinity to positive infinity. It is the kernel function of the Fourier transform.

[0021] Furthermore, in step S2, the abnormal data exceeding the normal range in the multi-source data is cleaned and filled according to data preprocessing principles to obtain the current information. The method is as follows:

[0022] The obtained current information refers to the current sequence after missing data filling and anomaly removal. Smoothing and predictive corrections are performed using higher-order differences and spline reconstruction. The optimal parameters are then obtained by minimizing the reconstruction error. The expression is:

[0023] ,

[0024] In the formula, Indicates the first Individual battery clusters at a given time point The estimated current value, Indicates from arrive The summation symbol, Indicates the first spline coefficients Indicates the first The B-spline basis functions at time point The value on, Represents the parameter vector Find the minimum value. Indicates from a point in time arrive Sum all data points. This indicates the use of B-spline basis functions. Fitting using linear combinations This represents the square of the fitting error. It is the regularization coefficient. This represents the square integral of the second derivative of the estimated current function.

[0025] Furthermore, in step S3, the method for determining the battery cluster's operating status based on the current information and outputting the determination result is as follows:

[0026] The battery cluster operating conditions include discharging when the current is greater than 5A, charging when the current is less than -5A, and resting when the current is in the range of -5A to 5A. An operating condition function is constructed based on this current information, expressed as:

[0027] ,

[0028] In the formula, Indicates the first Individual battery clusters at a given time point The operating conditions Indicates the first Individual battery clusters at a given time point The current value after filtering. Indicates the upper threshold of the current. Indicates the lower threshold of the current. This indicates that the current is within this range.

[0029] Furthermore, in step S3, the method for determining the battery cluster's operating status based on the current information and outputting the determination result is as follows:

[0030] The output judgment result introduces a fuzzy boundary function, defines the state membership degree using the maximum membership principle, models the uncertainty of the state boundary, and introduces a hidden Markov model, using a forward-backward algorithm to calculate the time step. The maximum posterior probability of a state is calculated using a sliding window. Smooth the current state, expression:

[0031] ,

[0032] In the formula, Indicates a time window, Indicates the window width. Indicates from the current time point Forward At a certain point in time, Indicate category At the current time point The moving average score, Indicates a point in time The state belongs to the category The score, This indicates the calculation of the mean of the sliding window. To express summation, Indicates the first Individual battery clusters at a given time point The state judgment result Represents all possible state categories Find the moving average in the middle. The largest possible state.

[0033] Further, in step S4, a data segment within the charging / discharging time period is obtained based on the judgment result. Principal component analysis is performed on the feature values ​​collected from the data segment to extract data features. The method for extracting data features is as follows:

[0034] The acquisition of data segments within the charging / discharging time period specifically involves selecting data segments during a single charging / discharging cycle when the relay is in normal condition. The expression is:

[0035] ,

[0036] In the formula, Indicates a point in time A feature vector was collected. Indicates the first Individual battery clusters at a given time point The current value after filtering. Indicates the first Individual battery clusters at a given time point The terminal voltage value, Indicates a point in time The temperature at the fuse, Indicates a point in time Temperature of the connector's insertion part Indicates a point in time The ambient temperature is set based on the total current and the measured voltage. Individual battery clusters at a given time point The current value and the measured voltage are respectively and Then its terminal voltage model is:

[0037] ,

[0038] In the formula, Indicates the first Individual battery clusters at a given time point The terminal voltage value, It is open-circuit voltage. Indicates the first Individual battery clusters at a given time point The equivalent internal resistance value was initially estimated by fitting the OCV-SoC curve, based on... For cases of slow change, least squares fitting is used to fit multi-point data over a period of time to calculate the internal resistance of each battery cluster in the energy storage system as a characteristic value. The expression is:

[0039] ,

[0040] In the formula, This means finding the minimum internal resistance value. , Indicates a time period All time points Summation, Indicates the first All sampling time points during the charge / discharge process Indicates the first Open-circuit voltage of each battery cluster This indicates all sampling time points Perform summation.

[0041] Further, in step S4, a data segment within the charging / discharging time period is obtained based on the judgment result. Principal component analysis is performed on the feature values ​​collected from the data segment to extract data features. The method for extracting data features is as follows:

[0042] The charge and discharge data are downsampled within a specific time period. For high-frequency disturbance signals, low-pass filtering and resampling are applied, expressed as follows:

[0043] ,

[0044] In the formula, Indicates a point in time The value after low-pass filtering and resampling. Indicates the window width. This indicates calculating the average value. Indicates from Until Adding each item together, Indicates the original signal at time point The value on, Indicates a point in time The data feature sequence after downsampling. The downsampling sampling factor indicates the sampling interval. Each time point retains a single value. Indicates a point in time A feature vector was collected. Indicates only in yes Only when the time is an integer multiple of the time, will the corresponding time point be retained. A feature vector is collected, and the temperature of each measuring point in the high-voltage box over a period of time is statistically analyzed and used as the feature value. Principal component analysis is performed based on the collected feature value to reduce the dimensionality of the data features and obtain data features containing more important information.

[0045] Further, in step S5, the data features are input into the anomaly identification algorithm to calculate the degree of anomaly in each high-voltage box within the battery cluster, identify high-voltage boxes with abnormal slopes, and perform anomaly warnings.

[0046] The calculation involves assessing the anomaly level of each high-voltage cell within the battery cluster, calculating the Mahalanobis distance between each sample and other samples, and applying adjustable parameters. , obtain samples With the Distance of each sample point The distance from P is less than or equal to All the points are of Calculate the distance to the neighborhood for any sample and The reachable distance, i.e. and The maximum value is used to calculate the Local Anomaly Factor (LOF) score for each sample based on the reachability distance. If the LOF value exceeds the set threshold, the local anomaly factor is considered an anomaly point, and a high-voltage box alarm can be triggered.

[0047] Further, in step S5, the data features are input into the anomaly identification algorithm to calculate the degree of anomaly in each high-voltage box within the battery cluster, identify high-voltage boxes with abnormal slopes, and perform anomaly warnings.

[0048] The anomaly warning system calculates the anomalous factors of each sample within the battery cluster, obtains the trend of anomalous factor changes over a period of time, predicts the slope of the anomalous factors, identifies whether there are high-voltage cells with abnormal slopes, and detects anomalous outliers that persist. The high-voltage box will issue an early warning of abnormality in a timely manner.

[0049] Beneficial effects

[0050] Compared with known public technologies, the technical solution provided by this invention has the following beneficial effects:

[0051] In use, this invention adaptively and rapidly identifies anomalies within the high-voltage box using an anomaly detection algorithm. It also provides early warnings of anomaly trends, scores data features, and quantifies the degree of anomaly of sample points within the energy storage system, improving the efficiency of anomaly quantification. The anomaly detection algorithm adaptively locates anomalies within the high-voltage box of the energy storage system's battery clusters, and then adaptively identifies outlier sample points without requiring prior knowledge. Furthermore, based on the slope of the scoring results, it quantifies the anomaly trends of each battery cluster and provides timely warnings, thereby improving the safety and performance of the energy storage system. Attached Figure Description

[0052] Figure 1 This is a flowchart of a method for identifying abnormalities in a high-voltage box of an energy storage system according to the present invention. Detailed Implementation

[0053] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0054] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but includes other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0055] The present invention will now be described in further detail with reference to the accompanying drawings:

[0056] Example:

[0057] like Figure 1 As shown, the present invention provides a method for identifying anomalies in the high-voltage box of an energy storage system, including the following steps: S1, acquiring multi-source data from each battery cluster in the energy storage system;

[0058] Furthermore, in step S1, the method for obtaining multi-source data within each battery cluster of the energy storage system is as follows:

[0059] The multi-source data includes timestamps within each battery cluster, total current, main positive and main negative terminals within the high-voltage box, pre-charge relay status, fuse temperature within the high-voltage box, connector insertion temperature, and measured voltage within the high-voltage box for each battery cluster. Set the collection period At every moment Collect the following multi-source data ,expression:

[0060] ,

[0061] In the formula, Indicates the first Individual battery clusters at a given time point The collected multi-source data set Indicates the first The timestamp of each sampling time point It is a timestamp indicating the number The specific time of each data collection Indicates the first Individual battery clusters in time Total current, Indicates the first Individual battery clusters in time The state of the main positive relay, Indicates the first Individual battery clusters in time The status of the main negative relay, Indicates the first Individual battery clusters in time The status of the pre-charged relay, Indicates the first Individual battery clusters in time The temperature of the fuse, Indicates the first Individual battery clusters in time The temperature of the connector insertion point Indicates the first Individual battery clusters in time The voltage measured inside the high-voltage box.

[0062] In this embodiment, multi-source data acquisition covers multiple dimensions such as electrical status, temperature safety, and relay logic, providing assurance for multi-angle status perception, facilitating the improvement of system data integrity, and ensuring consistent data time base for subsequent feature extraction and model building, thereby improving the accuracy of time series analysis. The acquired raw data can directly support the input of processes such as operating condition identification, feature extraction, and fault prediction.

[0063] S2. Based on the data preprocessing principles, the abnormal data exceeding the normal range in the multi-source data are cleaned and filled to obtain current information;

[0064] Furthermore, in step S2, the abnormal data exceeding the normal range in the multi-source data is cleaned and filled according to data preprocessing principles to obtain the current information. The method is as follows:

[0065] The cleaning and filling process involves removing duplicate data, using a set mapping function to uniquely process the timestamp set, filling missing values ​​with adjacent values ​​using a linear interpolation filling mechanism, and constructing a joint Gaussian model for abnormal current, temperature, and voltage data caused by any acquisition or communication issues. Mahalanobis distance is defined to detect anomalies, and Fourier transform is used to detect frequency drift for filtering. The expression is:

[0066] ,

[0067] In the formula, This represents the Fourier transform operator. Indicates the first Individual battery clusters in time The current signal, The integral symbol represents the integral from negative infinity to positive infinity. It is the kernel function of the Fourier transform.

[0068] Furthermore, in step S2, the abnormal data exceeding the normal range in the multi-source data is cleaned and filled according to data preprocessing principles to obtain the current information. The method is as follows:

[0069] The obtained current information refers to the current sequence after missing data filling and anomaly removal. Smoothing and predictive corrections are performed using higher-order differences and spline reconstruction. The optimal parameters are then obtained by minimizing the reconstruction error. The expression is:

[0070] ,

[0071] In the formula, Indicates the first Individual battery clusters at a given time point The estimated current value, Indicates from arrive The summation symbol, Indicates the first spline coefficients Indicates the first The B-spline basis functions at time point The value on, Represents the parameter vector Find the minimum value. Indicates from a point in time arrive Sum all data points. This indicates the use of B-spline basis functions. Fitting using linear combinations This represents the square of the fitting error. It is the regularization coefficient. This represents the square integral of the second derivative of the estimated current function.

[0072] In this embodiment, a systematic data cleaning and imputation method is adopted to address data quality issues such as duplicate values, missing items, and abrupt anomalies in multi-source data. This method obtains a high-quality current information sequence as the basis for subsequent feature extraction and fault identification, effectively eliminating redundancy and noise, and improving data reliability. The combination of Gaussian modeling and Mahalanobis distance detection makes anomaly detection more robust and interpretable. By combining Fourier spectrum analysis with spline fitting, the core pattern of current trend changes is preserved, avoiding information loss.

[0073] S3. Determine the operating status of the battery cluster based on the current information and output the determination result;

[0074] Furthermore, in step S3, the method for determining the battery cluster's operating status based on the current information and outputting the determination result is as follows:

[0075] The battery cluster operating conditions include discharging when the current is greater than 5A, charging when the current is less than -5A, and resting when the current is in the range of -5A to 5A. An operating condition function is constructed based on this current information, expressed as:

[0076] ,

[0077] In the formula, Indicates the first Individual battery clusters at a given time point The operating conditions Indicates the first Individual battery clusters at a given time point The current value after filtering. Indicates the upper threshold of the current. Indicates the lower threshold of the current. This indicates that the current is within this range.

[0078] Furthermore, in step S3, the method for determining the battery cluster's operating status based on the current information and outputting the determination result is as follows:

[0079] The output judgment result introduces a fuzzy boundary function, defines the state membership degree using the maximum membership principle, models the uncertainty of the state boundary, and introduces a hidden Markov model, using a forward-backward algorithm to calculate the time step. The maximum posterior probability of a state is calculated using a sliding window. Smooth the current state, expression:

[0080] ,

[0081] In the formula, Indicates a time window, Indicates the window width. Indicates from the current time point Forward At a certain point in time, Indicate category At the current time point The moving average score, Indicates a point in time The state belongs to the category The score, This indicates the calculation of the mean of the sliding window. To express summation, Indicates the first Individual battery clusters at a given time point The state judgment result Represents all possible state categories Find the moving average in the middle. The largest possible state.

[0082] In this embodiment, based on the continuous collection and analysis of battery cluster current information, the system automatically identifies the current battery cluster as being in one of three typical operating conditions: charging, discharging, or resting. This serves as the basis for subsequent data filtering and fault warning. By combining multiple mechanisms such as filtering, fuzzy membership, hidden Markov models, and sliding windows, the system greatly suppresses short-term misjudgments and frequent fluctuations. Through the sliding window mechanism, the system adapts to online calculations, and the forward-backward algorithm can quickly obtain the posterior probability of the state, making it suitable for edge deployment.

[0083] S4. Based on the judgment result, obtain the data segment within the charging and discharging time period, perform principal component analysis on the feature values ​​collected from the data segment, and extract data features;

[0084] Further, in step S4, a data segment within the charging / discharging time period is obtained based on the judgment result. Principal component analysis is performed on the feature values ​​collected from the data segment to extract data features. The method for extracting data features is as follows:

[0085] The acquisition of data segments within the charging / discharging time period specifically involves selecting data segments during a single charging / discharging cycle when the relay is in normal condition. The expression is:

[0086] ,

[0087] In the formula, Indicates a point in time A feature vector was collected. Indicates the first Individual battery clusters at a given time point The current value after filtering. Indicates the first Individual battery clusters at a given time point The terminal voltage value, Indicates a point in time The temperature at the fuse, Indicates a point in time Temperature of the connector's insertion part Indicates a point in time The ambient temperature is set based on the total current and the measured voltage. Individual battery clusters at a given time point The current value and the measured voltage are respectively and Then its terminal voltage model is:

[0088] ,

[0089] In the formula, Indicates the first Individual battery clusters at a given time point The terminal voltage value, It is open-circuit voltage. Indicates the first Individual battery clusters at a given time point The equivalent internal resistance value was initially estimated by fitting the OCV-SoC curve, based on... For cases of slow change, least squares fitting is used to fit multi-point data over a period of time to calculate the internal resistance of each battery cluster in the energy storage system as a characteristic value. The expression is:

[0090] ,

[0091] In the formula, This means finding the minimum internal resistance value. , Indicates a time period All time points Summation, Indicates the first All sampling time points during the charge / discharge process Indicates the first Open-circuit voltage of each battery cluster This indicates all sampling time points Perform summation.

[0092] Further, in step S4, a data segment within the charging / discharging time period is obtained based on the judgment result. Principal component analysis is performed on the feature values ​​collected from the data segment to extract data features. The method for extracting data features is as follows:

[0093] The charge and discharge data are downsampled within a specific time period. For high-frequency disturbance signals, low-pass filtering and resampling are applied, expressed as follows:

[0094] ,

[0095] In the formula, Indicates a point in time The value after low-pass filtering and resampling. Indicates the window width. This indicates calculating the average value. Indicates from Until Adding each item together, Indicates the original signal at time point The value on, Indicates a point in time The data feature sequence after downsampling. The downsampling sampling factor indicates the sampling interval. Each time point retains a single value. Indicates a point in time A feature vector was collected. Indicates only in yes Only when the time is an integer multiple of the time, will the corresponding time point be retained. A feature vector is collected, and the temperature of each measuring point in the high-voltage box over a period of time is statistically analyzed and used as the feature value. Principal component analysis is performed based on the collected feature value to reduce the dimensionality of the data features and obtain data features containing more important information.

[0096] In this embodiment, based on the judgment results of the battery cluster in charging, discharging, or resting states, the data segment with normal relay status during a single charge / discharge process is selected as the analysis object. Feature extraction and principal component analysis are performed. Composite features are constructed through internal resistance modeling, temperature statistics, and filtering and noise reduction, which fully reflect the operating conditions of the battery cluster and facilitates enhanced feature expression capabilities. Principal component analysis effectively compresses the high-dimensional feature space, reduces the computational pressure of subsequent algorithms, and facilitates reduced computational complexity. The dimensionality-reduced features are more consistent with the assumptions of the LOF anomaly detection algorithm, effectively reducing false alarms and improving anomaly identification accuracy.

[0097] S5. Input the data features into the anomaly identification algorithm to calculate the degree of anomaly of each high-voltage box in the battery cluster, identify high-voltage boxes with abnormal slopes, and issue an anomaly warning.

[0098] Further, in step S5, the data features are input into the anomaly identification algorithm to calculate the degree of anomaly in each high-voltage box within the battery cluster, identify high-voltage boxes with abnormal slopes, and perform anomaly warnings.

[0099] The calculation involves assessing the anomaly level of each high-voltage cell within the battery cluster, calculating the Mahalanobis distance between each sample and other samples, and applying adjustable parameters. , obtain samples With the Distance of each sample point The distance from P is less than or equal to All the points are of Calculate the distance to the neighborhood for any sample and The reachable distance, i.e. and The maximum value is used to calculate the Local Anomaly Factor (LOF) score for each sample based on the reachability distance. If the LOF value exceeds the set threshold, the local anomaly factor is considered an anomaly point, and a high-voltage box alarm can be triggered.

[0100] Further, in step S5, the data features are input into the anomaly identification algorithm to calculate the degree of anomaly in each high-voltage box within the battery cluster, identify high-voltage boxes with abnormal slopes, and perform anomaly warnings.

[0101] The anomaly warning system calculates the anomalous factors of each sample within the battery cluster, obtains the trend of anomalous factor changes over a period of time, predicts the slope of the anomalous factors, identifies whether there are high-voltage cells with abnormal slopes, and detects anomalous outliers that persist. The high-voltage box will issue an early warning of abnormality in a timely manner.

[0102] In this embodiment, anomaly detection and trend prediction are performed on the preprocessed data features to identify high-voltage boxes that may fail. This helps to increase the lead time for high-voltage box failure detection, reduces the risk of cluster downtime caused by failure to handle abnormal high-voltage boxes in time, and enhances the system's robustness to complex operating conditions and data disturbances. It is especially suitable for energy storage system scenarios with high sampling frequency and large data dimensions.

[0103] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying abnormality of a high-voltage box of an energy storage system, characterized in that, The method comprises the following steps: S1, acquiring multi-source data in each battery cluster in the energy storage system; S2, cleaning and filling abnormal data exceeding the normal range in the multi-source data according to a data preprocessing principle to obtain current information; S3, judging the working condition state of the battery cluster according to the current information and outputting a judgment result; S4, acquiring a data segment in a charging and discharging time period according to the judgment result, performing principal component analysis on characteristic values collected from the data segment, and extracting data characteristics; S5, inputting the data characteristics into an abnormality recognition algorithm, calculating the abnormality degree of each high-voltage box in the battery cluster, identifying a slope-abnormal high-voltage box, and performing abnormality early warning.

2. The method of claim 1, wherein, In step S1, the method for acquiring multi-source data in each battery cluster in the energy storage system is: The multi-source data includes time stamp, total current, main positive, main negative, pre-charge relay status, fuse temperature, connector pin temperature, measured voltage in high voltage box, for each battery cluster , set the collection period , at each time , collect the following multi-source data , expression: ， In the formula, represents the battery cluster at the time point collected multi-source data set, represents the time stamp of the sampling time point, is the time stamp representing the specific time of the collected data, represents the total current of the battery cluster at the time , represents the state of the main positive relay of the battery cluster at the time , represents the state of the main negative relay of the battery cluster at the time , represents the state of the pre-charging relay of the battery cluster at the time , represents the temperature of the fuse of the battery cluster at the time , represents the temperature of the connector plug of the battery cluster at the time , represents the voltage measured in the high-voltage box of the battery cluster at the time .

3. The method of claim 2, wherein, In step S2, the method for cleaning and filling abnormal data exceeding the normal range in the multi-source data according to a data preprocessing principle to obtain current information is: The cleaning and filling, deleting duplicate data, using a set mapping function to uniquely process the timestamp set, filling missing values according to adjacent values, using a linear interpolation filling mechanism, constructing a joint Gaussian model for abnormal data of current, temperature and voltage caused by any one of collection or communication problems, defining Mahalanobis distance to detect abnormal points, filtering by Fourier transform to detect frequency drift, expression: ， In the formula, This represents the Fourier transform operator. Indicates the first Individual battery clusters in time The current signal, The integral symbol represents the integral from negative infinity to positive infinity. It is the kernel function of the Fourier transform.

4. The method of claim 3, wherein, In step S2, the method for cleaning and filling abnormal data exceeding the normal range in the multi-source data according to a data preprocessing principle to obtain current information is: The obtained current information is for the current sequence after missing data filling and abnormal data elimination , and the optimal parameters are obtained by minimizing the reconstruction error through smoothing and predictive correction using high-order difference and spline reconstruction. ， wherein denotes the estimated current value of the th battery cluster at time point , denotes the summation symbol from to , denotes the th spline coefficient, denotes the value of the th B-spline basis function at time point , denotes the minimization over the parameter vector , denotes the summation over all data points from time point to , denotes the fitting using a linear combination of B-spline basis functions , denotes the squared error of the fit, is a regularization coefficient, denotes the squared integral of the second derivative of the estimated current function.

5. The method of claim 4, wherein, In step S3, the method for judging the working condition state of the battery cluster according to the current information and outputting a judgment result is: The working condition state of the battery cluster includes discharging when the current is greater than 5A, charging when the current is less than -5A, and standing when the current is in [-5, 5]A, constructing a working condition state function according to the current information, expression: ， wherein represents the operating state of the battery cluster at time point , represents the operating state of the battery cluster at time point filtered current value, represents the upper threshold value of the current, represents the lower threshold value of the current, represents that the current is in this interval.

6. The method of claim 5, wherein, In step S3, the method for judging the working condition state of the battery cluster according to the current information and outputting a judgment result is: The output judgment result introduces a fuzzy boundary function, defines state membership degree, adopts the maximum membership degree principle, carries out uncertain modeling of state boundary, introduces a hidden Markov model, uses a forward-backward algorithm to calculate the maximum posterior probability of state at time , adopts a sliding window to smooth the current state, and the expression is: ， wherein denotes a time window, denotes the window width, denotes the current time point the preceding time points, denotes the class the sliding average score at the current time point , denotes the score for the state belonging to the class at the time point , denotes the calculation of the mean value of the sliding window, denotes the summation, denotes the state determination result of the battery cluster at the time point , denotes the finding of the state among all possible state classes that maximizes the sliding mean value.

7. The method of claim 6, wherein, In step S4, the method for acquiring a data segment in a charging and discharging time period according to the judgment result, performing principal component analysis on characteristic values collected from the data segment, and extracting data characteristics is: The method for acquiring a data segment in a charging and discharging time period is specifically selecting a data segment of a single charging / discharging and a normally working relay, expression: ， In the formula, represents the temperature at the time point of the fuse, represents the temperature at the time point of the connector insertion part, represents the temperature at the time point of the environment, represents the voltage value of the first battery cluster at the time point , represents the current value of the first battery cluster at the time point , represents the voltage value of the first battery cluster at the time point , represents the current value of the first battery cluster at the time point , represents the voltage value of the first battery cluster at the time point , represents the current value of the first battery cluster at the time point , ， In the formula, represents the terminal voltage value of the mth battery cluster at the time point t, is the open circuit voltage, represents the equivalent internal resistance value of the mth battery cluster at the time point t, The initial estimate is fitted by the OCV-SoC curve, and the internal resistance value of each battery cluster in the energy storage system is calculated as a characteristic value according to the slow change of The least square fitting is used to fit the multi-point data in a period of time, and the expression is:​​​ ， wherein denotes finding the minimized internal resistance value , denotes summing over all time points in the time period , denotes all sampling time points in the th charge / discharge process, denotes the open circuit voltage of the th battery cluster, denotes summing over all sampling time points .

8. The method of claim 7, wherein, In step S4, the method for acquiring a data segment in a charging and discharging time period according to the judgment result, performing principal component analysis on characteristic values collected from the data segment, and extracting data characteristics is: The charge and discharge data are down-sampled, and the time period In the foregoing, the high-frequency disturbance signal is filtered by a low-pass filter and resampled, and the expression is: ， In the formula, represents the value after low-pass filtering + resampling processing at time point represents the window width, represents the average value, represents from to add each item together, represents the value of the original signal at time point represents the data feature sequence after down-sampling processing at time point is the down-sampling sampling factor, which means that every time point is kept, represents a feature vector collected at time point represents only when is an integer multiple of , the corresponding feature vector collected at time point is kept, the temperature of each measuring point in the high-pressure box within a period of time is counted as a feature value, principal component analysis is performed according to the collected feature value, the data features are reduced, and data features containing more important information are obtained.​​​​ 9. The method of claim 8, wherein, In step S5, the method for inputting the data characteristics into an abnormality recognition algorithm, calculating the abnormality degree of each high-voltage box in the battery cluster, identifying a slope-abnormal high-voltage box, and performing abnormality early warning is: The calculation involves assessing the anomaly level of each high-voltage cell within the battery cluster, calculating the Mahalanobis distance between each sample and other samples, and applying adjustable parameters. , obtain samples With the Distance of each sample point The distance from P is less than or equal to All the points are of Calculate the distance to the neighborhood for any sample and The reachable distance, i.e. and The maximum value is used to calculate the Local Anomaly Factor (LOF) score for each sample based on the reachability distance. If the LOF value exceeds the set threshold, the local anomaly factor is considered an anomaly point, and a high-voltage box alarm can be triggered.

10. The method of claim 8, wherein, In step S5, the method for inputting the data characteristics into an abnormality recognition algorithm, calculating the abnormality degree of each high-voltage box in the battery cluster, identifying a slope-abnormal high-voltage box, and performing abnormality early warning is: The abnormal early warning calculates the abnormal factor of each sample in the battery cluster, obtains the change trend of the abnormal factor in the past period, performs slope prediction on the abnormal factor, identifies whether there is a slope abnormal high-pressure box, and performs high-pressure box abnormal early warning when there is an abnormal outlier and continuous time.

Images